This probably only applies to other folks who are running yarn as a different user than default.
I upgraded from HDP 1.3 -> 2.0 via manual RPM upgrade. One thing I will mention to hopefully save others some time is that all of my yarn nodes were coming up as “UNHEALTHY”. The reason stated was this via the UI:
ERROR check containerexecutor, OK: disks ok,
It turns out, that health check was returning ERROR because the perms on /usr/lib/hadoop-yarn/bin/container-executor looked like this:
—Sr-s— 1 root yarn 40K Oct 17 17:40 /usr/lib/hadoop-yarn/bin/container-executor
but needed to be this:
—Sr-s— 1 root hadoop 40K Oct 17 17:40 /usr/lib/hadoop-yarn/bin/container-executor
Simple chown/chmod fixed the problem and the cluster righted itself.